AGENT SKILL

supervise

Supervise an epic or complex task end-to-end within a container session -- assemble workflows via /reify, coordinate subagents with incremental git checkpoints, deliver in-session artifacts and pull requests, and queue out-of-session follow-ups. Use for autonomous epic execution, high-level task supervision, and multi-stage delivery. Exclude for single atomic tasks already reified on the graph (use pull).

License 许可
MIT
最近更新
2026-11-15
Stars
★ 5
Forks
3

01 前言 Front Matter

name
supervise
description
Supervise an epic or complex task end-to-end within a container session -- assemble workflows via /reify, coordinate subagents with incremental git checkpoints, deliver in-session artifacts and pull requests, and queue out-of-session follow-ups. Use for autonomous epic execution, high-level task supervision, and multi-stage delivery. Exclude for single atomic tasks already reified on the graph (use pull).

03 SKILL.md 正文

/supervise: End-to-End Epic Execution and Supervision

Supervise a repository-scoped epic task to completion inside a container session. The lead executor (james) runs this skill to reify workflows onto a combined task body, coordinate subagent reviews and fan-out, commit and push incrementally to prevent pre-emption loss, deliver in-session outputs, and wire follow-up tasks for out-of-session work.

0. Immediate Claim & Live PKB Persistence

  1. Claim the Task: Claim the assigned task immediately via pkb_claim_task. If invoked on an untracked directive without a task ID, search existing work and create a repository-scoped tracking task via /q, then claim it.
  2. Continuous Persistence: Update the PKB task record continuously after each completed stage or subagent chunk so progress and state survive session pre-emption.

1. In-Session Workflow Assembly (/reify)

Check the claimed task body before composing:

  • Fast-Path: If the task already carries a composed workflow (## Instructions and ## Appendix: Sources) or requires invoking a single skill (invoke /<skill>), skip redundant re-composition and proceed to planning.
  • Workflow Composition: Otherwise, invoke /reify in-session to compose the workflow from the three-tier template library onto the single task body. Adapt multi-stage workflows to execute wholly within the current container session.

2. Plan & Subagent Dialectical Review

  1. Draft Execution Plan: Draft the execution plan onto the PKB task record. When modifying agent behavior, run an enforcement-pyramid options pass against specs/enforcement/enforcement.md before adopting new constraints.
  2. Parallel Dialectical Review: Dispatch a proportionate parallel subagent review across specialist lenses covering strategic direction and knowledge-base alignment, axiom and rule compliance, and qualitative output excellence. Append a null argdown block (asks: none / [no claim]) to claim-free briefs.
  3. Verify and Resolve: Check returned subagent arguments per ida:premise-check before acting on or relaying findings. Cull completed helpers, resolve blocking findings, and apply uncontroversial fixes before implementation fan-out.

3. Model-Calibrated Subagent Fan-Out & Incremental Push Checkpoints

  1. Calibrated Delegation: Fan out disjoint work chunks in parallel to appropriately sized subagent models:
    • High-capability models: Complex architecture, specifications, and multi-constraint synthesis.
    • Fast lightweight models: Focused file edits, template updates, mechanical sweeps, and index synchronization.
    • Provide complete inputs and target artifact paths in Turn-1 briefs, and instruct subagents to return their final ida:premise-check report cleanly on exit without duplicate mid-turn status messages.
  2. Incremental Checkpoint Invariant: After every chunk of subagent work:
    • Verify handbacks via ida:premise-check.
    • Run formatting and linters on modified files.
    • Stage and commit modified files strictly scrubbed of private identifiers, PKB task/note/session IDs, Task: trailers, personal names, and private hostnames (data-boundaries).
    • Run git push immediately to the upstream branch so work survives container pre-emption.
    • Update the PKB task record with incremental progress.

4. In-Session Delivery & Out-of-Session Follow-Ups

  1. In-Session Verification: Complete all in-session deliverables: full test suites, linters, builds, and an independent final diff review across strategic, compliance, and quality lenses.
  2. Pull Request: Open and ready a single cohesive pull request against the base branch, attaching scrubbed review receipts.
  3. Out-of-Session Follow-Up Tasks: Deliver everything executable in-session. Create and wire a repository-scoped follow-up task on the PKB graph (via /q placement with serves value-chain links and needs prerequisite edges) only when work explicitly requires:
    • A different repository (strictly one task per repository).
    • A different time or external wait condition.
    • A workflow-mandated separate external agent or session (such as closed-loop user-channel delivery).

5. Handover (/dump)

Invoke /dump to record the ## Report argument proof, pr_url, and branch. Release the task as done when all in-session deliverables and pull requests are complete (and any out-of-session work is queued on follow-up tasks), or release as partial with blocking reasons if in-session deliverables remain unfinished.

04 同仓库更多 Skills

nicsuzor
nicsuzor/academicOps

supervise

Supervise an epic or complex task end-to-end within a container session -- assemble workflows via /reify, coordinate subagents with incremental git checkpoints, deliver in-session artifacts and pull requests, and queue out-of-session follow-ups. Use for autonomous epic execution, high-level task supervision, and multi-stage delivery. Exclude for single atomic tasks already reified on the graph (use pull).

★ 5
nicsuzor
nicsuzor/academicOps

supervise

Supervise an epic or complex task end-to-end within a container session -- assemble workflows via /reify, coordinate subagents with incremental git checkpoints, deliver in-session artifacts and pull requests, and queue out-of-session follow-ups. Use for autonomous epic execution, high-level task supervision, and multi-stage delivery. Exclude for single atomic tasks already reified on the graph (use pull).

★ 5
nicsuzor
nicsuzor/academicOps

supervise

Supervise an epic or complex task end-to-end within a container session -- assemble workflows via /reify, coordinate subagents with incremental git checkpoints, deliver in-session artifacts and pull requests, and queue out-of-session follow-ups. Use for autonomous epic execution, high-level task supervision, and multi-stage delivery. Exclude for single atomic tasks already reified on the graph (use pull).

★ 5
nicsuzor
nicsuzor/academicOps

supervise

Supervise an epic or complex task end-to-end within a container session -- assemble workflows via /reify, coordinate subagents with incremental git checkpoints, deliver in-session artifacts and pull requests, and queue out-of-session follow-ups. Use for autonomous epic execution, high-level task supervision, and multi-stage delivery. Exclude for single atomic tasks already reified on the graph (use pull).

★ 5
nicsuzor
nicsuzor/academicOps

madr

Draft Markdown Architectural Decision Records (MADR) capturing context, considered options, decision outcomes, and consequences. Use when recording an architectural decision, documenting a technical choice, or evaluating design alternatives. Exclude for declarative target-state specifications (use specs) or task planning (use reify).

★ 5
nicsuzor
nicsuzor/academicOps

madr

Draft Markdown Architectural Decision Records (MADR) capturing context, considered options, decision outcomes, and consequences. Use when recording an architectural decision, documenting a technical choice, or evaluating design alternatives. Exclude for declarative target-state specifications (use specs) or task planning (use reify).

★ 5